statsmodels.regression.linear_model.OLS.hessian#

OLS.hessian(params, scale=None)[source]#

Evaluate the Hessian function at a given point.

Parameters:

paramsarray_like: The parameter vector at which the Hessian is computed.
scalefloat, optional: If None, return the Hessian of the profile (concentrated) log-likelihood (profiled over the scale parameter), else return the Hessian of the log-likelihood using the given scale value.

Returns:

ndarray: The Hessian matrix.

Notes

With scale=None the Hessian of the profile log-likelihood is -n / 2 * (2 X'X / SSR - g g' / SSR**2), where SSR is the sum of squared residuals and g is its gradient with respect to params.

Changed in version 0.15.1: The Hessian with scale=None was corrected. The first term was previously too small by a factor of 2.